Hardware–Software Co-Design for Decimal Multiplication

Riaz-Ul-Haque Mian,Michihiro Shintani,Michiko Inoue

doi:10.3390/computers10020017

Riaz-Ul-Haque Mian, Michihiro Shintani + Show 1 more

Open Access

https://doi.org/10.3390/computers10020017

Copy DOI

Abstract

Decimal arithmetic using software is slow for very large-scale applications. On the other hand, when hardware is employed, extra area overhead is required. A balanced strategy can overcome both issues. Our proposed methods are compliant with the IEEE 754-2008 standard for decimal floating-point arithmetic and combinations of software and hardware. In our methods, software with some area-efficient decimal component (hardware) is used to design the multiplication process. Analysis in a RISC-V-based integrated co-design evaluation framework reveals that the proposed methods provide several Pareto points for decimal multiplication solutions. The total execution process is sped up by 1.43× to 2.37× compared with a full software solution. In addition, 7–97% less hardware is required compared with an area-efficient full hardware solution.

Highlights

As this study considers a co-design methodology where the delay improvement in hardware is not a considerable amount compared to the other part, we adopt the area efficient binary-coded decimal (BCD)-Carry-lookahead adder (CLA) [31]
The proposed methods are implemented on an RISC-V based integrated framework for the software–hardware co-design proposed in Reference [21]
The operations corresponding to the hardware part in the proposed methods are integrated as custom instructions in a Rocket Chip [33,34], and they are executed in an accelerator through the Rocket Custom Coprocessor (RoCC) interface

Summary

Introduction

Decimal arithmetic is very important for banking, commercial, and financial transactions. It is widely used in scientific applications. A study by IBM has indicated that more than half of the numerical data in a commercial database are stored in decimal format [1]. Decimal arithmetic in binary logic causes some errors that change the actual value of the exact result. This is completely unacceptable in financial or other critical scientific operations.

Objectives

Methods

Results

Discussion

Conclusion